stakesense-mcp

Model Context Protocol server exposing predictive Solana validator quality scoring to Claude Desktop, Claude Code, Cursor, and any MCP-compatible LLM agent.

What it gives an agent

Eleven tools and five resources, all backed by the public stakesense REST API:

Tools

| Tool | What it answers | |---|---| | get_validator_score | Latest pillar + composite scores for a vote pubkey, with metadata + recent history | | recommend_top_validators | "Where should I stake N SOL?" — ranked list with reasoning | | list_validators | Browse top of any pillar (composite / downtime / mev_tax / decentralization) | | get_validator_history | "Is this validator getting better or worse?" — score trajectory | | get_validator_rank | Where a validator ranks + gap to top-10/50 (operator-side framing) | | get_decentralization_report | Network-wide Nakamoto coefficient + ASN/data-center/country breakdowns | | get_concentration_by | Top clusters along one axis (data_center / asn / country) | | get_network_stats | Global summary (averages, latest epoch, validator count) | | get_recent_anomalies | What changed in the validator set this week (MEV jumps, delinquency, score moves) | | simulate_migration | What-if rebalance: pass before/after allocations, get composite + concentration deltas | | health_check | API liveness + last refresh timestamp |

Resources (read-once context)

| URI | Contents | |---|---| | stakesense://methodology | Long-form methodology paper (markdown) | | stakesense://model-card | Technical model documentation (markdown) | | stakesense://manifest | Open-data export manifest (JSON, CC-BY 4.0) | | stakesense://network-stats | Latest network averages + Nakamoto coefficient | | stakesense://decentralization | Concentration breakdown snapshot |

Install

Claude Desktop / Claude Code

claude mcp add stakesense -- npx stakesense-mcp

That's it — restart Claude and ask: "Which Solana validators have the lowest predicted downtime risk?"

Cursor / Windsurf

Add to your MCP config (Cursor: Settings → Features → MCP, or ~/.cursor/mcp.json):

{
  "mcpServers": {
    "stakesense": {
      "command": "npx",
      "args": ["-y", "stakesense-mcp"]
    }
  }
}

Manual / from source

git clone https://github.com/mikejohnkurkeyerian-eng/stakesense
cd stakesense/mcp
pnpm install
pnpm build
node dist/index.js   # speaks MCP over stdio

Configuration

| Env var | Default | Purpose | |---|---|---| | STAKESENSE_API_BASE | https://stakesense.onrender.com | Override to point at a self-hosted stakesense backend |

Example prompts

Once installed, your agent has predictive Solana validator data on tap:

• "List the top 5 validators by composite score and explain why each one ranks high."
• "I have 100 SOL to stake — pick three validators for a balanced risk profile and tell me why."
• "What's Solana's current Nakamoto coefficient, and which data centers host the most validators?"
• "Show me the prediction history for <vote_pubkey> over the last 30 days. Is it trending up or down?"
• "Compare the decentralization-score distribution across the top 50 validators."

Data freshness

Stakesense refreshes twice daily (04:17 + 16:17 UTC) via a GitHub Actions cron. The health_check tool reports last_update_epoch and last_prediction_date so your agent can confirm freshness before answering time-sensitive questions.

License

MIT. The data the server exposes is published under CC-BY 4.0 — please attribute stakesense if you reuse it.

Bug? Idea?

Open an issue on the main repo. PRs welcome.